>
> For a bigger database (34386 documents), I encounter a performance drop in
> ftinfex use : a few seconds in 11.9 against 17 minutes with the last
> release.
> Words with a few hits respond quickly but when hits are numerous (here
> around 4000), it takes many minutes (here 17 minutes).
>
I guess it would take longer… but probably not much as in your case.
Basically, the enforceindex option is a lax convenience workaround for
cases in which the optimizer is not smart enough to rewrite the expression
for index access. However, to ensure that the results are correct, and
match the input documents, several steps need to be performed that are
otherwise done once at compile time. Specifically, an expression like…
$documents//text()[. contains text 'X']
…is rewritten to something like…
ft:search('db', 'X')[some $doc in $documents satisfies $doc is
ancestor::document-node()]
In your case, it’s probably better to directly use ft:search. There are
several approaches how to ensure that the results are contained in specific
documents of a database. One approach is to match the database paths of the
document and the result nodes:
declare function local:getHit($documents, $word) {
let $db := db:name(head($documents))
let $paths := $documents ! db:path(.)
return ft:search($db, $word)[db:path(.) = $paths]
};
Hope this helps,
Christian
